Artificial intelligence communications agent

Abstract

Systems and methods for artificial intelligence communications agents are disclosed. Implementations relate to capturing individual agent's behaviors and modelling them in artificial intelligence (AI) learning models so that the agent's behavior can be easily replicated. Some implementations further relate to systems and methods for capturing human-computer interactions (HCl) performed by agents and using robotic process automation (RPA) to automate tasks that would otherwise require human interaction. The combination of AI learning models and RPA are used to provide artificial intelligence communications agents capable of responding to a variety of topics of conversation over a variety of communication mediums.

What is claimed is: 
     
       1. A method performed by a computing device for modeling agent behavior, the method comprising:
 accessing transcripts of communications between customers and agents, each communication comprising one or more query/response pairs of a query by a customer and a response by an agent, each transcript having a set of associated computer interactions performed by the agent based on the communication of the transcript and correlated to the query/response pairs; 
 generating topic clusters of the transcripts based on similarity of topics of the transcripts; 
 for each topic cluster of transcripts,
 for each transcript in that topic
 cluster, for each 
 query/response pair,
 generating a feature vector for the query of that query/response pair; 
 labeling the feature vector based on the response of that query/response pair and any computer interactions correlated to that query/response pair; and 
 
 
 training a model for that topic cluster using the label feature vectors generated from the transcripts of that topic cluster, 
 wherein each trained model for a topic cluster identifies responses and computer interactions when provided a query associated with a communication relating to a topic that is similar to the topic of that topic cluster so that the identified responses can automatically be provided and the identified computer interactions can be translated to generate one or more robotic process automation (RPA) tasks that can be automatically performed. 
 
 
     
     
       2. The method of  claim 1  further comprising: receiving a policy including a set of requirements corresponding to the transcript, wherein the model is further trained to comply with the policy. 
     
     
       3. The method of  claim 1  further comprising: applying dimension reduction to the feature vector for the query. 
     
     
       4. The method of  claim 1 , wherein the transcripts are derived from communications over a first communications medium, and
 wherein the model is configured to respond to communications over the first communications medium. 
 
     
     
       5. The method of  claim 4 , wherein the communications medium is one of telephonic communications, email communications, text chat communications, social media, smart device, or Internet of Things device communications. 
     
     
       6. The method of  claim 1 , wherein the computer interactions performed by the agent are received from a human-computer interaction recorder installed on a computer terminal used by the agent. 
     
     
       7. The method of  claim 1 , wherein the feature vector for the query is generated using term frequency-inverse document frequency (tf-IDF) or term frequency- proportional document frequency (tf-PDF). 
     
     
       8. The method of  claim 1  further comprising: computing a ranking metric value for each trained model, wherein the trained models for each topic cluster are ranked according to the computed ranking metric value. 
     
     
       9. A method performed by a robotic process automation module at a computing device for providing recommendations of agent behavior, the method comprising:
 receiving a query by a customer; 
 generating a feature vector for the received query; 
 identifying a trained model having been trained using transcripts of communications between customers and agents, wherein the trained model identifies responses and robotic process automation (RPA) tasks when provided one or more queries; 
 generating a response and a set of RPA tasks in response to the received query by applying the identified trained model to the generated feature vector for the received query; 
 executing, by one or more robotic process agents, at least one RPA task in the generated set of RPA tasks; and 
 sending the generated response to be provided to the customer. 
 
     
     
       10. The method of  claim 9 , further comprising adding the generated set of RPA tasks to a data structure of RPA tasks, wherein the data structure of RPA tasks is located at a second computing device that is communicatively coupled to the computing device. 
     
     
       11. The method of  claim 9 , wherein the generated set of RPA tasks includes at least one task generated by a customer service agent. 
     
     
       12. The method of  claim 9 , wherein the trained model has been further trained to comply with a policy, and
 wherein the policy includes a set of requirements corresponding to an organization. 
 
     
     
       13. The method of  claim 12 , wherein the feature vector for the received query is generated based on the policy. 
     
     
       14. The method of  claim 9 , wherein at least one RPA task in the generated set of RPA tasks corresponds to a computer interaction that could be performed by an agent in response to the query. 
     
     
       15. The method of  claim 9  further comprising:
 identifying a communication medium via which the query is received, 
 wherein the trained model has been further trained to respond to queries received via the communication medium. 
 
     
     
       16. The method of  claim 15 , wherein the feature vector for the received query is generated based on the communication medium. 
     
     
       17. A method performed by a computing device for modeling agent behavior, the method comprising:
 accessing transcripts of communications between customers and agents, each communication comprising one or more query/response pairs of a query by a customer and a response by an agent, each transcript having a set of associated computer interactions performed by the agent based on the communication of the transcript and correlated to the query/response pairs; 
 generating topic clusters of the transcripts based on similarity of topics of the transcripts; 
 for each topic cluster of transcripts,
 for each transcript in that topic
 cluster, for each 
 query/response pair,
 generating a feature vector for the query of that query/response pair; 
 labeling the feature vector based on the response of that query/response pair and any computer interactions correlated to that query/response pair; 
 
 
 training a model for that topic cluster using the label feature vectors generated from the transcripts of that topic cluster, 
 wherein each trained model for a topic cluster identifies responses and robotic process automation (RPA) tasks when provided a query associated with a communication relating to a topic that is similar to the topic of that topic cluster so that the identified responses can automatically be provided and the identified RPA tasks can be automatically performed; 
 
 receiving a query by a customer; 
 generating a feature vector for the received query; 
 accessing a trained model having been trained using transcripts of communications between customers and agents, wherein the trained model identifies responses and RPA tasks when provided one or more queries; 
 generating a response and a set of RPA tasks in response to the received query by applying the accessed trained model to the generated feature vector for the received query; 
 executing, by one or more robotic process agents, at least one RPA task in the generated set of RPA tasks; and 
 sending the generated response to be provided to the customer. 
 
     
     
       18. The method of  claim 17 , wherein the trained model has been further trained to comply with a policy, and
 wherein the policy includes a set of requirements corresponding to an organization. 
 
     
     
       19. The method of  claim 17  further comprising:
 identifying a communication medium via which the query is received, 
 wherein the trained model has been further trained to respond to queries received via the communication medium. 
 
     
     
       20. The method of  claim 17  further comprising:
 computing a ranking metric value for each model, 
 wherein the models for each topic cluster are ranked according to the computed ranking metric value.